Stable histone deacetylase complexes distinguished by the presence of SANT domain proteins CoREST/kiaa0071 and Mta-L1.

Human histone deacetylases I (HDAC1) and II (HDAC2) are homologous proteins (84% identity) that catalyze release of acetyl groups from modified N-terminal lysines of core histones. Histone deacetylation is correlated with both transient and persistent states of transcriptional inactivity (i.e. silencing) in many eukaryotes. In this study, we analyzed complexes containing HDAC1 and HDAC2 to identify the proteins most stably associated with these deacetylases. Complex cI (9.5 S) contained transcriptional corepressor CoREST/kiaa0071 and a protein homologous to FAD-dependent oxidoreductases, kiaa0601. Complex cII (15 S) contained >/=15 proteins, including CHD3/4 (Mi-2), Mta-L1, RbAp48/46, and MBD3, characteristic of vertebrate nucleosome-remodeling complexes. Under native conditions, cI and cII may contain HDAC1, HDAC2 or both; these can be dissociated to cI and cII core complexes containing only HDAC1 or HDAC2. The (m)CpG-binding protein MBD2 was associated only with the HDAC1 cII core complex. A model is proposed in which HDAC1 core complexes can be targeted to methylated DNA via MBD2 with recruitment of HDAC2 occurring through formation of HDAC1/2 cII dimers. We note that the cI component CoREST/kiaa0071 and the cII component Mta-L1 share a region of homology that includes a SANT domain; this domain may play a role in complex assembly.

Core histones are subject to reversible acetylation at selected lysine residues in their N-terminal domains. Histone acetylation is correlated with transcriptional activity or competence such that hyperacetylation within chromatin domains favors transcriptional competency, whereas hypoacetylation favors transcriptional silencing. Acetylation neutralizes positive charges on lysine ⑀-amino groups and may alter nucleosome conformation by weakening interactions of the N-terminal domains with DNA or adjacent nucleosomes. In addition, acetylation at specific residues may create or abolish binding sites for non-histone chromosomal proteins (1)(2)(3).
Acetylation states are maintained by the coordinate activity of histone acetyltransferases and histone deacetylases (HDACs). 1 Histone acetyltransferases are oxidoreductases, transferring an acetyl group from acetyl-CoA, whereas HDACs remove acetyl groups by hydrolysis (4,5). Most HDACs belong to a superfamily of zinc metalloenzymes with a conserved ϳ380-residue catalytic domain (6,7). The class I HDACs (ϳ500 residues in length) are the best characterized with respect to function. These HDACs are involved in maintaining both shortterm (induction/repression) and long-term (epigenetic) patterns of gene activity (8 -10). HDAC mutants of Rpd3 in Saccharomyces cerevisiae and Clr6 in Schizosaccharomyces pombe affect telomeric silencing (11,12); similarly, HDAC mutants in Drosophila, RPD3 (13), and Caenorhabditis elegans, HDA-1 (14), affect silencing of developmentally regulated genes.
Three class I HDACs have been identified in human. HDAC1 (482 residues) (15) and HDAC2 (488 residues) (16) share 84% identity, whereas HDAC3 (428 residues) (17,18) is more divergent, sharing 51% identity with HDAC1/2. It is unclear whether HDAC1 and HDAC2 have distinct functions, but they appear to play complementary roles in transcriptional repression. For example, in GH 3 pituitary cells, thyroid-stimulating hormone-␤ expression is negatively regulated by the thyroid hormone triiodothyronine. This triiodothyronine-dependent regulation requires HDAC activity, as indicated by its sensitivity to the inhibitor trichostatin A. An HDAC1 multiprotein complex binds constitutively to a negative regulatory element in the thyroid-stimulating hormone-␤ promoter. However, chromatin remodeling and transcriptional repression at the thyroid-stimulating hormone-␤ promoter require (in addition) triiodothyronine-dependent recruitment of an HDAC2 complex to the negative regulatory element (19).
Immunopurification of either HDAC1 or HDAC2 from human cell lines typically yields multiprotein complexes containing both HDAC1 and HDAC2. These include HDAC complexes containing transcriptional corepressor mSin3A (20 -22) and chromatin-remodeling ATPase Mi-2 (NURD complex) (23)(24)(25)(26)(27)(28). To gain further insight into the mechanism by which HDAC1 and HDAC2 complexes may be independently recruited to specific sites in chromatin, we have analyzed core complexes containing either HDAC1 or HDAC2. Two major HDAC1/2 complexes were identified in HeLa cells: cI is a novel complex containing transcriptional corepressor CoREST/kiaa0071 (29) and protein kiaa0601, inferred to be an FAD-dependent enzyme; cII is similar to the Mi-2 remodeling complex. Under native conditions, cI and cII may contain HDAC1, HDAC2, or both, but these can be dissociated by a mild chaotrope to core complexes containing only HDAC1 or HDAC2. This behavior suggests a general model for interaction between HDAC1 and HDAC2 via dimerization of cI-or cII-specific components.
We identified a potential HDAC-targeting factor, m CpGbinding protein MBD2 (30), as an integral component of the HDAC1 cII complex. MBD2 was not detected in HDAC2 cII complexes, implying that targeting of HDAC2 may occur via dimerization with HDAC1. Finally, we note that a region of sequence similarity exists between cI component CoREST/ kiaa0071 and cII component Mta-L1 (31), containing SANT domains (32). SANT domain proteins are also components of HDAC3 core complexes (33) and may play a general role in HDAC complex assembly.

EXPERIMENTAL PROCEDURES
Cell Culture and Transduction-HeLa S3 cells were grown in suspension culture as previously described (34). Cells were transduced with a retroviral construct (pOZ) expressing a bicistronic mRNA encoding FLAG-tagged HDAC1 protein linked to an interleukin-2 receptor subunit surface marker, and the transduced population was purified by repeated cycles of affinity cell sorting (34). A cDNA construct encoding human HDAC1 with a C-terminal FLAG tag, used to construct the retroviral vector, was generously provided by Dr. Stuart L. Schreiber. A cDNA encoding human HDAC2 (accession number NP_001518) was isolated by performing polymerase chain reaction using gene-specific primers located in the 5Ј-and 3Ј-untranslated regions with a CLON-TECH Marathon cDNA library; this cDNA clone was used to prepare an expression construct with a C-terminal FLAG tag (see above). For detection of FAD, 2.8 ϫ 10 8 cells were radiolabeled with 1 mCi of [ 3 H]riboflavin (25 Ci/mmol; American Radiolabeled Chemicals) in 350 ml of Dulbecco's modified Eagle's medium (minus riboflavin) for 16 h prior to harvest.
Isolation of HDAC1/2 Complexes-In view of reports that histone deacetylases may be resistant to extraction from isolated nuclei (39), we employed a sequential extraction procedure to facilitate recovery of soluble complexes from the transduced cells. For preparation of nuclei, cells were washed twice in phosphate-buffered saline, resuspended in hypotonic buffer (20 mM MES (pH 6.5), 10 mM NaCl, 1.5 mM MgCl 2 , and 1 mM DTT), and lysed by Dounce homogenization (40), yielding nuclei and cytosol. The nuclear pellet was washed in isotonic sucrose buffer (STM: 250 mM sucrose, 50 mM Tris (pH 7.4), 5 mM MgCl 2 , and 1 mM DTT), yielding STM wash, and the nuclear envelope was removed by addition of 1% Triton X-100, yielding Triton wash. Nuclei were sequentially extracted with low magnesium buffer (10 mM Tris (pH 7.4), 0.2 mM MgCl 2 , and 1 mM DTT), yielding low magnesium wash, and then with low magnesium buffer plus 0.5 M NaCl (41,42), yielding high salt extract and residual nuclei. All buffers contained Complete protease inhibitor (Roche Molecular Biochemicals).
Most of HDAC1/2 was obtained in the high salt extract (see Fig. 1A). The high salt extract was dialyzed in ion-exchange chromatography buffer (20 mM Tris (pH 8.0), 100 mM NaCl, 1 M urea, 1 mM DTT, 10% glycerol, and 0.005% CHAPS), centrifuged at 27,000 ϫ g, and applied to a Q-Sepharose column (Amersham Pharmacia Biotech) equilibrated with chromatography buffer. The column was eluted with a gradient of 100 -750 mM NaCl, and the HDAC1-containing fractions (ϳ250 -350 mM NaCl) were pooled and dialyzed in immunopurification buffer (20 mM Tris (pH 8.0), 150 mM NaCl, 5 mM MgCl 2 , 10% glycerol, and 0.1% Tween 20). The Q-Sepharose pool was mixed with either anti-FLAG or anti-HDAC1 antibody-agarose for 3 h on a rotator, packed into a column, and washed in ϳ50 volumes of immunopurification buffer. Bound proteins were eluted from anti-FLAG antibody-agarose by incubation with immunopurification buffer containing FLAG peptide (50 g/ml) or from anti-HDAC1 antibody-agarose with 100 mM glycine (pH 2.5) and 0.1% Tween 20. (Note that removal of urea by dialysis was required to efficiently bind the FLAG epitope; endogenous HDAC1 could be immunopurified without dialysis. The same proteins co-immunopurified with endogenous HDAC1 in the presence or absence of 1 M urea.) To resolve HDAC1/2 complexes, 200 l of anti-FLAG eluate was centrifuged in a 3.7-ml 10 -35% glycerol gradient for 5.5 h at 55,000 rpm using an SW 60Ti rotor (Beckman Instruments) (43). Twenty 200-l fractions were collected. Marker proteins (ovalbumin (4.54 S), aldolase (7.3 S), catalase (11.3 S), and thyroglobulin (19 S)) were centrifuged in parallel gradients.
Histone Deacetylase Enzyme Assay-Histone deacetylase activity was measured using human core histones radiolabeled with [ 3 H]acetate in vivo as substrate (44). Aliquots (1-10 l) were incubated with 10 g of [ 3 H]histone in 100 l of assay buffer (20 mM Tris (pH 8.0), 75 mM NaCl, and 1 mM DTT) for 20 min at 37°C. The reaction was quenched by addition of 20 l of 12 N HCl and extracted with 0.5 ml of ethyl acetate. Acetate release was measured by liquid scintillation counting of the ethyl acetate (organic) phase. As a specificity control, release of [ 3 H]acetate was inhibited by either 0.1 M trichostatin A or 10 mM sodium butyrate.
Identification of Proteins by Mass Spectrometry-Protein identification by mass spectrometry was carried out as described (34) with minor modifications. An electrospray ion trap mass spectrometer (LCQ, Finnigan MAT, San Jose, CA) coupled on line with a Magic 2002 capillary HPLC (Michrom BioResources, Auburn, CA) was used to acquire tandem mass spectra. A 0.1 ϫ 50-mm Magic MS C 18 column (5-m particle diameter, 200-Å pore size) with mobile phases A (methanol/water/acetic acid, 5:95:1) and B (methanol/water/acetic acid, 85:15:1) was used with a gradient of 2-98% mobile phase B over 2.5 min, followed by 98% mobile phase B for 2 min. Specific protein bands from Coomassie Blue-stained SDS-polyacrylamide gels were excised, destained, and digested with trypsin, and the resulting peptides were extracted. 20 -50% of the sample was used to obtain liquid chromatography/tandem mass spectra. The tandem mass spectra were used to search the compiled NCBI nonredundant protein and expressed sequence tag data bases with the program PepFrag to identify the proteins.

Purification of HDAC1/2 Complexes-HDAC1 and HDAC2
were extracted from isolated HeLa cell nuclei with 0.5 M NaCl (high salt extract) (see Fig. 1A (HS) and "Experimental Procedures"). Endogenous HDAC1 and HDAC2 were isolated using specific immunoadsorbents recognizing their (nonconserved) C-terminal 16 residues. When HDAC1 was isolated directly from the high salt extract under non-stringent conditions, some HDAC2 co-immunopurified with it ( Fig. 1B), as in previous studies (22,25,27,45,46). We examined chaotropic agents as a means to fully dissociate HDAC1 from HDAC2. Histone deacetylase enzymatic activity was stable indefinitely in 1 M urea, but was irreversibly lost at higher concentrations (2-4 M). HDAC1 and HDAC2 were fully dissociated by Q-Sepharose chromatography in 1 M urea. The proteins coeluted in the Q-Sepharose pool (0.25-0.35 M NaCl), but bound independently to the immunoadsorbents (Fig. 1B).
Expression of Epitope-tagged HDAC1/2 in HeLa Cells-To facilitate physical characterization of purified, native HDAC1 and HDAC2 multiprotein complexes, we expressed recombinant proteins with a C-terminal FLAG epitope tag (HDAC1-F and HDAC2-F) in HeLa S3 cells. Cell lines with stable expression of either HDAC1-F or HDAC2-F were obtained by transduction with a bicistronic retroviral vector and repeated cycles of magnetic affinity cell sorting (see "Experimental Procedures"). Endogenous and recombinant HDAC1/2 proteins expressed in these cells were quantified by Western blotting using affinity-purified antibody specific to the C terminus of either HDAC1 or HDAC2, respectively, and mouse anti-FLAG mAb M2. Levels of endogenous and recombinant HDAC1 proteins were compared using a baculovirus-expressed HDAC1 protein with an N-terminal FLAG tag as a standard; approximately equal amounts of endogenous and recombinant HDAC1 proteins were expressed in the transduced cells (Fig. 1A).
Endogenous and FLAG-tagged HDAC1 proteins were isolated from the Q-Sepharose pool (see above) using anti-HDAC1 or anti-FLAG immunoadsorbents, respectively. The polypeptides copurifying with either endogenous or FLAG-tagged HDAC1 from the Q-Sepharose pool were compared by SDS-PAGE (Fig. 2). The polypeptide composition of the anti-FLAG eluate was generally similar to that of the anti-HDAC1 eluate, except for the presence of several additional bands. These additional bands were also present in the control anti-FLAG eluate, indicating that they were contaminants; they were removed by further purification of the HDAC1-F complexes (see below).
Sedimentation Analysis of HDAC1 and HDAC2 Complexes-The variable staining intensities of the HDAC1-associated polypeptides suggested that they might represent a mixture of different multiprotein complexes. We employed velocity sedimentation in a glycerol density gradient to separate HDAC1-F complexes differing in size and shape. Histone deacetylase enzyme activity sedimented as a 15 S peak with a 9.5 S shoulder (Fig. 3, upper panel). The distribution of HDAC1-F by Western blotting (Fig. 3, lower panel) was similar to the activity profile, indicating similar specific activities for all sedimenting species of HDAC1.
Different polypeptides co-sedimented with HDAC1-F in the 9.5 S and 15 S regions of the gradient (Fig. 3, middle panel), indicating that these fractions represent different core complexes, designated cI (9.5 S) and cII (15 S). The polypeptide components of cI (fractions 6 and 7) and cII (fractions 9 and 10) were analyzed by mass spectrometry (see above). The p63 band in both cI and cII contained HDAC1 and kiaa0071. The p115 band in cI contained kiaa0601. cII contained the remodeling complex components Mi-2 (p240), Mta-L1 (p78), RbAp48/46 (p55/p53), and MBD3 (p36/p33) and, in addition, MBD2 (p50/ p30). Additional bands were identified in cII: p70 is a human homolog of the Xenopus laevis HDAC-associated zinc finger protein p66 (24); p59 and p38 represent truncated forms of Mta-L1. The narrow width of the 15 S peak (same as the sedimentation standards) suggests that cII represents a single multiprotein complex. (This idea is further supported by immunopurification studies described below.) The variation in staining intensity of the bands could be due either to different stoichiometries (i.e. 3-5 mol of p240/p78/p63/p55/p53 per mol of p50/p36/p33) or to differences in the amount of Coomassie Blue dye bound per mol of each protein.
To obtain specific probes for cI and cII, affinity-purified antipeptide antibodies were prepared against the C-terminal 16 residues of kiaa0601, kiaa0071, Mta-L1, and MBD2. kiaa0601 and kiaa0071 were found predominantly in cI, whereas Mta-L1 and MBD2 were found exclusively in cII (Fig. 3, lower panel). The data from Coomassie Blue staining and Western blotting suggest that cI is a multiprotein complex composed of kiaa0601 (p115), kiaa0071 (p63), HDAC1 (p63), and p37, which is consistent with the predicted molecular mass for a 9.5 S complex, ϳ200 kDa. The predicted molecular mass for a 15 S complex, ϳ400 kDa, would appear to be an underestimate for cII. Assuming only 1 mol of each visible band in the multiprotein complex would indicate a molecular mass of at least 1 MDa. , and residual nuclei (Nuc) prepared as described under "Experimental Procedures" were resolved by SDS-PAGE; proteins were visualized by Western blotting using rabbit anti-HDAC1 antibody and mouse anti-FLAG mAb M2. The quantitation standard was a recombinant HDAC1 protein with an N-terminal FLAG tag expressed in Sf9 cells. B, immunopurification of endogenous HDAC1. High salt extract (load (Ld)) was applied either to affinity-purified rabbit anti-HDAC1 antibody coupled to protein A-agarose (␣HDAC1) or to control agarose containing preimmune IgG (Control). Unbound protein was collected (flow-through (FT)), and the agarose was extensively washed and then eluted with 0.1 M glycine (pH 2.5) (eluate (Elt)). Alternatively, the extract was chromatographed on Q-Sepharose (Q-S), and a pool of fractions containing HDAC1 was applied to the immunoadsorbents. HDAC1 and HDAC2 in these fractions were detected by Western blotting. However, sedimentation coefficients are a function of both the molecular mass and frictional ratio, the latter being related to particle asymmetry (43). In view of the wide size range of the proteins in cII, it is likely to have a greater frictional ratio than the sedimentation standards; thus, 400 kDa would represent only a minimum estimate.
Endogenous HDAC1 and HDAC2 Are in Distinct cI and cII Core Complexes-To determine whether endogenous HDAC1 and HDAC2 in normal (untransduced) HeLa cells exist as distinct cI and cII core complexes, we employed antibody adsorbents specific for cI (anti-kiaa0601 and anti-kiaa0071) and cII (anti-Mta-L1 and anti-MBD2) components. Binding was performed using the native high salt extract (no urea treatment) under non-stringent conditions (see "Experimental Procedures"). After extensive washing, the adsorbents were eluted with pH 2.5 buffer, and the eluates were analyzed by SDS-PAGE (Fig. 5A). As expected, all adsorbents (except the control) bound HDAC1. (Protein recovery from the anti-MBD2 adsorbent was relatively low.) Also as predicted, the cI components kiaa0601 and kiaa0071 bound to anti-HDAC1, anti-kiaa0601, and anti-kiaa0071 adsorbents, but not to anti-Mta-L1 and anti-MBD2 adsorbents; conversely, the cII components Mta-L1 and MBD2 bound to both anti-Mta-L1 and anti-MBD2 adsorbents, but not to anti-kiaa0601 and anti-kiaa0071 adsorbents. This demonstrates that cI and cII are distinct HDAC1 complexes in the native high salt extract. Blotting indicated that mSin3A was not associated with either cI or cII.
HDAC2 was also bound by both cI-and cII-specific adsorbents (Fig. 5A), as might be expected, since some HDAC2 is associated with HDAC1 (Fig. 1). To examine whether HDAC2 exists as cI or cII complexes independently of HDAC1, the fraction of endogenous HDAC2 not associated with HDAC1 ( Fig. 1) was isolated from the anti-HDAC1 flow-through (i.e. depleted of HDAC1) using an anti-HDAC2 adsorbent. Both cI components kiaa0071 and kiaa0601 and the cII component Mta-L1 co-immunopurified with HDAC2 (Fig. 5A). However, MBD2 was not detected and appears not be associated with endogenous HDAC2 cII, consistent with our analysis of FLAGtagged HDAC2 (Fig. 4).
Further insight into the composition and potential significance of cI and cII can be gleaned from examination of the polypeptide composition of the eluates (Fig. 5B). The anti-kiaa0071 eluate is similar to the cI region of the glycerol gradient, except for the additional p61 band (which is characteristic of endogenous HDAC1) (Fig. 2), supporting our interpretation that kiaa0601, kiaa0071, HDAC1, and p37 compose a multiprotein complex. The absence of any major polypeptides in the anti-kiaa0071 eluate besides the cI components indicates that cI represents the quantitatively major FIG. 3. Sedimentation analysis of FLAG-tagged HDAC1 complexes in a glycerol density gradient. The anti-FLAG eluate (see Fig. 2) was applied to a 10 -35% glycerol gradient and centrifuged at 55,000 rpm for 5.5 h using an SW 60Ti rotor. Sedimentation standards (ovalbumin (3.5 S), aldolase (7.3 S), catalase (11.3 S), and thyroglobulin (19 S)) were centrifuged in parallel gradients. Twenty 0.2-ml fractions were collected. Aliquots (2 l) were analyzed for histone deacetylase enzyme activity (upper panel); fractions representing 9.5 S cI and 15 S cII are indicated. Total proteins in each fraction were resolved by SDS-PAGE and visualized by Coomassie Blue staining (middle panel). The Coomassie Blue-stained bands referred to under "Results" are indicated (q). The apparent molecular masses for these bands are indicated on the left (cI) and right (cII). In the lower panels are shown the results from Western blotting for distribution of cI and cII components in the gradient: FLAG-tagged HDAC1 (anti-FLAG mAb) and endogenous kiaa0601, kiaa0071, Mta-L1, and MBD2 (see "Results").

FIG. 4. Sedimentation analysis of FLAG-tagged HDAC2 complexes in a glycerol density gradient.
The anti-FLAG eluate was centrifuged in a glycerol gradient as described in the legend to Fig. 3. Aliquots were analyzed for histone deacetylase enzyme activity (upper panel). Total proteins in each fraction were resolved by SDS-PAGE (lower panel). The apparent molecular masses for the major components are indicated on the right (see "Results").
form of kiaa0071 in the high salt extract. The anti-kiaa0601 eluate is similar to the anti-kiaa0071 eluate, except for a few additional polypeptides, notably p140 and p160. These polypeptides did not appear to cross-react with the anti-kiaa0601 antibody in Western blotting; we favor the interpretation that they are components of other multiprotein complexes containing kiaa0601, but not kiaa0071 or HDAC1/2.
When the polypeptide composition of the anti-Mta-L1 eluate was examined, it proved to be similar to the cII region of the gradients (Figs. 3 and 4). The absence of any other major polypeptides in the anti-Mta-L1 eluate besides the cII components indicates that cII represents the quantitatively major form of Mta-L1 in the high salt extract. The anti-MBD2 eluate is similar to the anti-Mta-L1 eluate. Despite the large difference in staining intensity between Mta-L1 (p78) and MBD2 (p50), these polypeptides appeared in a fixed ratio in all preparations containing cII, supporting our interpretation that cII represents a single molecular species (see above). Overall, our analysis has demonstrated that the majority of HDAC1 and HDAC2 in HeLa cells exist in two distinct types of multiprotein complexes, cI and cII, which may contain HDAC1, HDAC2, or both. cI and cII can be dissociated to core complexes containing only HDAC1 or HDAC2 in association with other enzyme and targeting factors. We propose a general model for interaction between HDAC1 and HDAC2 via dimerization of cI-or cIIspecific components (see "Discussion" and Fig. 7).
Complex cI Contains Riboflavin-The cI component kiaa0601 exhibits sequence similarity to a superfamily of FADdependent enzymes (see Fig. 8). To examine whether kiaa0601 may contain FAD, HeLa cells were radiolabeled in vivo with [ 3 H]riboflavin (see "Experimental Procedures"). Riboflavin is converted to FMN and subsequently to FAD. The immunoadsorbents specific for cI (anti-kiaa0071 and anti-kiaa0601) both bound ϳ5% of the radiolabeled flavin in the high salt extract (Table I), whereas the anti-Mta-L1 and control adsorbents bound negligible amounts (Յ0.01%). Since kiaa0601 and kiaa0071 are the major proteins bound specifically by the cI adsorbents (Fig. 5B), it is reasonable to infer that kiaa0601 contains the radiolabeled flavin.
Subcellular Localization of cI and cII Components-To gain further insight into the possible role of cI and cII components as targeting factors for HDACs, we examined their subcellular localization in human skin fibroblasts. The cI components kiaa0071 and kiaa0601 were distributed similarly to HDAC1, in the nucleoplasm and excluded from the nucleoli (Fig. 6). The cII component Mta-L1 was localized to the nucleoplasm and was also detected in the juxtanuclear Golgi complex (47). MBD2 exhibited a punctate distribution in nuclei, differing from HDAC1 and Mta-L1. This may be explained by the fact that only a minor fraction of MBD2 (ϳ30%) was extractable with the HDACs in 0.5 M NaCl; possibly the majority of MBD2 is in different complexes, tightly bound to chromatin.

DISCUSSION
Two Distinct HDAC1/2 Core Complexes-HDAC1 and HDAC2 are closely related mammalian histone deacetylases that appear to mediate complementary functions in transcriptional regulation at specific sites in chromatin (see the Introduction). To gain further insight into the mechanisms by which HDAC1 and HDAC2 interact, we have analyzed multiprotein complexes from HeLa cell nuclei containing HDAC1 and HDAC2. Complex cI is composed of kiaa0601 (p116), kiaa0071 (p63), p37 (unidentified), and HDAC, apparently in an equimolar ratio, based on sedimentation behavior (see "Results" and Fig. 3). Complex cII is similar to vertebrate deacetylase nucleosome-remodeling (NURD) complexes (23)(24)(25)(26)(27)(28) in that it contains Mi-2 (p240), Mta-L1/Mta2 (p78), RbAp48/46 (p55/p53), and MBD3 (p36/p33) as well as a homolog of the Xenopus-specific component p66 (p70). Under native conditions, cI and cII may contain HDAC1, HDAC2, or both; however, mild dissociation by 1 M urea yields core complexes containing only HDAC1 or HDAC2. This behavior suggests a general model for interaction FIG. 5. Co-immunopurification of cI and cII components with endogenous HDAC1 and HDAC2. The high salt (HS) extract prepared from normal (untransduced) HeLa cells was applied to antibodyagarose containing either preimmune (Pre Imm) IgG (control) or affinity-purified antibody to HDAC1, Mta-L1, MBD2, kiaa0071, or kiaa0601. The flow-through from the anti-HDAC1 adsorbent (depleted of HDAC1) was applied to an anti-HDAC2 adsorbent to isolate HDAC2 not associated with HDAC1. Bound proteins eluted from the adsorbents (as described in the legend Fig. 2) were resolved by SDS-PAGE. A, Western blotting to identify HDAC1, HDAC2, mSin3A, Mta-L1, MBD2, kiaa0601, and kiaa0071 proteins in the eluates. B, polypeptide composition of the eluates visualized by Coomassie Blue staining. Control is an eluate from an antibody adsorbent to human Finb protein (see "Experimental Procedures"), which does not bind HDAC complexes. Solid bars indicate polypeptides copurifying with either cI (kiaa0601 and kiaa0071) or cII (Mta-L1 and MBD2) components. The bands referred to under "Results" are indicated (q); a cross-reacting protein in the anti-MBD2 eluate is shown (ϫ). The p63 and p61 bands contain HDAC1 (indicated on the right); and in addition, the p63 band contains kiaa0071 in the anti-kiaa0601, anti-kiaa0071, and anti-HDAC1 eluates (indicated on the left). between HDAC1 and HDAC2 via dimerization of cI-or cIIspecific components (Fig. 7). Association with specific folding partners appears to be required for HDAC1/2 function in vivo. When overexpressed in Sf9 insect cells, human HDAC1 and HDAC2 are largely insoluble and have little enzymatic activity, suggesting incorrect folding. 2 Coexpression with RbAp48/46 and Mta2 (also termed Mta-L1) is required to assemble soluble, enzymatically active HDAC complexes (23). The polypeptide complement required to assemble active HDAC complexes has been termed the "core complex" (23,25). The core complex components serve as folding partners for HDACs and, in addition, may mediate substrate binding, association with complementary enzymatic functions, and targeting of HDACs to specific sites in chromatin (e.g. promoters).
A requirement for HDAC-folding partners can also be inferred from the behavior of epitope-tagged HDAC1/2 stably expressed by our retroviral vector in transduced cells. The levels of expression of HDAC1/2 obtained using this system in mammalian cells appear to be strictly limited compared with expression in bacteria or insect cells. At most, we have been able to obtain only a 4 -5-fold excess of recombinant over endogenous HDAC1 (in normal mouse skin fibroblasts), and this was accompanied by a decreased level of endogenous HDAC1, amounting to only a 3-fold increase overall. The absence of significant amounts of free (monomeric) HDAC1 and HDAC2 implies that they must associate with specific folding partners for metabolic stability; presumably, excess HDAC1/2 is turned over and does not accumulate. Comparison of the relative abundance of cI-and cII-specific components associated with endogenous versus FLAG-tagged HDAC1 indicates that recombinant HDAC1 has a greater tendency to form cII complexes (Fig. 2). Similarly, FLAG-tagged HDAC2 formed only cII complexes (Fig. 4), although endogenous HDAC2 formed both cI and cII complexes (Fig. 5). This tendency to form cII complexes may be due to a greater availability of the cII components as folding partners or could be due to an effect of the FLAG tag.
HDAC1 cII contains additional polypeptides not present in HDAC2 cII (Figs. 4 and 5). One of these polypeptides was FIG. 8. Predicted amino acid sequence of protein kiaa0601 and similarity to maize polyamine oxidase. Shown is the amino acid sequence of protein kiaa0601, predicted from the DNA sequence of cDNA clone HG0838a (accession number AB011173) merged with expressed sequence tag clone DKFZp434I0321 (accession number AL042550), which extends HG0838a 5Ј by 57 base pairs. The merged sequence contains two in-frame stop codons located Ϫ132 and Ϫ150 base pairs from the first start codon in HG0838a, defining the initiation codon for kiaa0601 (position 1). The kiaa0601 amino acid sequence (K; accession number BAA25527) is aligned with similar sequences in human BAF170 (B; SWI3 homolog; accession number NP_003066) and maize polyamine oxidase (P; accession number CAA05249); identities are shaded. The three FAD and two substrate-binding domains (Sub1 and Sub2) in the polyamine oxidase crystal structure are indicated (substrate-binding region 1, omitted, is not similar to kiaa0601). Asterisks indicate residues contacting FAD in polyamine oxidase (51).

FIG. 6. Subcellular localization of cI and cII components.
Shown are the results from the immunofluorescent staining of human skin fibroblasts using affinity-purified antibodies to HDAC1, cI components kiaa0071 and kiaa0601, and cII components Mta-L1 and MBD2, followed by fluorescein isothiocyanate-conjugated anti-rabbit antibody. Arrows indicate localization of Mta-L1 to the juxtanuclear Golgi complex, in addition to the nucleoplasm (see "Results").
identified as the m CpG-binding protein MBD2 (30). MBD2 is a transcriptional corepressor (48) that binds specifically to m CpG residues in DNA and exhibits a preference for densely methylated regions. In vivo, MBD2 is localized to heterochromatic regions in mouse cells (30). MBD2 may direct Mi-2 remodeling complexes to methylated regions in chromatin, mediating transcriptional silencing (23,24). Based on our model (Fig. 7), HDAC1 cII could be recruited directly to methylated regions in DNA, but recruitment of HDAC2 to these sites would require heterodimerization of HDAC2 cII with HDAC1 cII.
Significance of Complex cI-Protein kiaa0071, also termed CoREST, binds to the RE1 silencing transcription factor (REST) in vitro and forms a complex with REST in the L6 skeletal muscle cell line (29). CoREST exhibits transcriptional corepressor activity in a Gal4 DNA-binding domain-mediated reporter assay, even in cells lacking REST. It is plausible that CoREST may function as a corepressor by recruiting HDAC1 to promoters, as has been proposed for MBD2 (see above) and other deacetylase-associated transcriptional repressors (48). REST is a transcriptional silencer of neural-specific genes in non-neural cells; it may serve to target cI complexes to promoters for neural-specific genes.
Protein kiaa0601 appears to belong to a superfamily of FADdependent enzymes (49, 50) based on sequence similarity (Fig.  8) and binding radiolabeled flavin (Table I). Conservation is strong at residues contacting FAD (Fig. 8) within the three FAD-binding domains of maize polyamine oxidase (51). Particularly significant is the conservation at Gly 57 and Lys 300 . These residues in polyamine oxidase position a water molecule at the site of hydrogen exchange (N-5) in the isoalloxazine ring and may represent the catalytic center. The N-terminal one-third of kiaa0601 resembles that of CoREST/kiaa0071 in having a block of alanines and also has similarity to conserved region 1 of the SWI3-like subunits of the SWI/SNF chromatin-remodeling complex (52). The fact that a large fraction of kiaa0601 exists in a complex with a histone deacetylase and a transcriptional corepressor (Fig. 5) suggests that kiaa0601 may be an enzyme with an activity complementary to that of the deacetylase in transcriptional silencing. The superfamily of FAD enzymes to which kiaa0601 apparently belongs includes a diverse group of oxidases and dehydrogenases involved mostly in biosynthetic/degradative pathways (49). In the context of cI, it seems plausible that kiaa0601 may be an oxidoreductase involved in covalent modification of chromatin constituents.
Role of Conserved Domains in HDAC Core Complex Assembly-The cI component kiaa0071 and the cII component Mta-L1 (Mta2) share sequence similarity in an ϳ200-residue segment that is tandemly repeated in kiaa0071 (Fig. 9). This region of similarity between kiaa0071 and Mta-L1 contains a SANT domain (32). Interestingly, the transcriptional corepressors SMRT and NCoR, which have recently been identified as components of HDAC3 core complexes (33), 3 are likewise SANT domain proteins, suggesting a general role for SANT domain proteins as folding partners for HDACs. The ϳ60-residue SANT domain (related to the Myb DNA-binding domain) is found in a variety of transcription factors, present in one to three copies, and may mediate protein-protein interactions. We further note that the arrangement of the two SANT domains in kiaa0071, separated by an ϳ100-residue spacer, is similar to that in SMRT and NCoR (29). This suggests that differing configurations of the conserved SANT domains in kiaa0071 and Mta-L1 may have a role in specific assembly of cI and cII core complexes.